Executive Summary

Learning—by brains or machines—is not a copyright-relevant act. This essay argues that treating training as presumptively infringing (rescued, sometimes, by fair use) mistakes the Constitutional purpose of copyright, misreads the statute’s fixation concepts, and further entrenches an unworkable permission culture. Drawing on doctrine (fixation, fair use, § 117’s “essential step”), technology history (player pianos to Google Books), and a step-by-step technical account of modern machine learning, I propose a simple rule: regulate outputs, not inputs; legalize learning.